In recent years, a technique for recognizing speech spoken by a person on a computer in accordance with a predetermined rule (so-called speech recognition technique) has been put into practical applications. Furthermore, a so-called client-server speech recognition system that shares speech recognition with a heavy load by an external speech recognition server having sufficient computer resources and performance upon implementing a speech recognition function on a less powerful portable terminal is used (Japanese Laid-Open Patent No. 7-222248).
On the other hand, a speech input client has been proposed. When the client-server speech recognition system is used, speech recognition that requires a large vocabulary and expert knowledge, and speech recognition that must be done after being connected to the network are made on the server side. However, speech recognition that requires a small vocabulary such as operation of the client side is done on the client to reduce the traffic on the network.
However, in the conventional speech recognition system, which speech recognition means in the client or server is used cannot be determined from input speech data. Furthermore, when a plurality of servers are connected to the speech recognition system, or when one server has a plurality of speech recognition means, and when these servers or speech recognition means can make speech recognition of different languages or that specialized for specific fields, the client cannot explicitly select and use the predetermined server or speech recognition means.